光学精密工程2025，Vol.33Issue(21)：3331-3342,12.DOI:10.37188/OPE.20253321.3331
基于边缘增强弱监督K-means的大口径望远镜拼接探测器点云分割
Edge-enhanced weak-supervised K-means point cloud segmentation for large-aperture telescope tiled detectors
摘要
Abstract
To satisfy the requirements for high-efficiency and high-precision flatness assessment of detec-tors used in large-aperture,wide-field telescopes,a K-means clustering segmentation method integrating edge-information constraints is proposed for mosaic detectors.Employed as a preprocessing step prior to flatness evaluation,the method reliably extracts regular structural regions and thereby enhances the accura-cy and stability of subsequent flatness-index computations.It is applicable to detector inspection across multiple stages and operating conditions.First,the structural characteristics of the spliced detector were analyzed to develop a dedicated point-cloud processing procedure.Edge-gap features were extracted and edge continuity was reinforced.An adaptive initialization of cluster centers within closed regions was im-plemented to avoid instability arising from random initialization.Furthermore,edge-penalty terms and height-anomaly detection mechanisms were incorporated into the K-means objective function.Through it-erative optimization,refined segmentation of individual detector point clouds was achieved.Experimental results indicate that centroid initialization within closed regions reduces the average number of iterations and removes the need for manual presetting of cluster counts.Compared with conventional methods,the incorporation of edge constraints and height-anomaly detection improves boundary-matching accuracy by more than 50%.The proposed method effectively segments point clouds of spliced detector datasets,of-fering an efficient segmentation approach for the development and deployment of spliced detectors in large-aperture telescopes.关键词
大口径光电望远镜/拼接探测器/点云分割/边缘检测/聚类Key words
large-aperture optical and infrared telescope/segmented detector/point cloud segmentation/edge detection/clustering分类
信息技术与安全科学引用本文复制引用
冯晓鹏,王之一,王建立,刘昌华,贾建禄,马爽,刘塔拉..基于边缘增强弱监督K-means的大口径望远镜拼接探测器点云分割[J].光学精密工程,2025,33(21):3331-3342,12.基金项目
国家自然科学基金资助项目(No.11973041) （No.11973041）
